techtubers
#programming
tender river
sage crag
get him off my screen
tender river
uneven pulsar
techtubers
are worst
noble zodiac
you can find youtube videos claiming X is bad for literally every topic
tender river
please do not get your opinions off social media
sage crag
you can find youtube videos claiming X is bad for literally every topic
especially for tech
tender river
on anything, really
uneven pulsar
you can find youtube videos claiming X is bad for literally every topic
X is shit
uneven pulsar
not X the platform
still shit platform
sage crag
especially for tech
guys deno is out- abandon v8, guys bun is out- abandon deno, guys
lament igloo
get him off my screen
oh? so his videos are not really true ab react and hes a react hater?
uneven pulsar
oh? so his videos are not really true ab react and hes a react hater?
did you hear what i said
sage crag
oh? so his videos are not really true ab react and hes a react hater?
these channels make money off sensationalism where there is no sensation 
lament igloo
i see
fierce nova
bro open your country/city job postings and see yourself, don't trust this shit
lament igloo
bro open your country/city job postings and see yourself, don't trust this shit
im in the philippines 
uneven pulsar
it became optional for me
tender river
oh? so his videos are not really true ab react and hes a react hater?
the reality is that react is widely used and facebook have done their best to make it work and have the monetary incentive to make it work as well as possible, react has went through multiple iterations to be as good as possible, and it's trusted by many
you can use svelte or vue or whatever you prefer
sage crag
there's always a problem to be fixed, but the truth is that enterprise devs still use the same languages and technologies from 10 years ago
because they are reliable
lament igloo
the reality is that react is widely used and facebook have done their best to ma...
hmmm i saw a video showing a react site rendering a lot of objects by just a single scroll π§ is it just bad code
sage crag
when it comes to things like c, they have been kicking around for literal decades and are still in use
uneven pulsar
the reality is that react is widely used and facebook have done their best to ma...
thats why they taught me react in my school
uneven pulsar
hmmm i saw a video showing a react site rendering a lot of objects by just a sin...
hey
lament igloo
i wanna clear my doubts in react so i can change my perspective of it
so forgive me if i sound stupid
tender river
hmmm i saw a video showing a react site rendering a lot of objects by just a sin...
react uses cached dom and lazy change propagation, if something is being redrawn thats very much just the devs making it so
uneven pulsar
you can use svelte or vue or whatever you prefer
svelte is acautly very nice fast and simple to use still it doesn't have that big economy like react but very good for small projects or big quick ones
sage crag
i wanna clear my doubts in react so i can change my perspective of it
move your doubts of react to doubting the web ecosystem entirely
its a steaming dumpsterfire
uneven pulsar
react uses cached dom and lazy change propagation, if something is being redrawn...
not to forget its fast
running on the jvm virtual machine
lament igloo
hmmm i saw a video showing a react site rendering a lot of objects by just a sin...
oh yeah this is the video https://www.youtube.com/watch?v=INLq9RPAYUw&pp=ygUKcmVhY3QgdGhlbw%3D%3D
by theo again lmfaoo
uneven pulsar
oh yeah this is the video https://www.youtube.com/watch?v=INLq9RPAYUw&pp=ygUKcmV...
what NO
its fast
noble zodiac
why does he look confused on every thumbnail
sage crag
why does he look confused on every thumbnail
he's confused
uneven pulsar
why does he look confused on every thumbnail
react fast
sage crag
can confirm
uneven pulsar
thanks
these people are the same people who post unemployed shit on twitter about AI hype and stuff
uneven pulsar
how do u explain this π€¨
the devs are stupid
sage crag
lament igloo
~~github devs are stupid? ~~
sage crag
the devs are stupid
the devs arent stupid, this is simply a mistake
uneven pulsar
the devs arent stupid, this is simply a mistake
probably
sage crag
probably<:Tomfoolery:1189430206595858462>
not probably
you cant assume that they are stupid based on this
dont call people stupid for trivial mistakes
noble zodiac
how do u explain this π€¨
thats completely irrelevant. No one looks at this and goes: NO, I will stop using github now, this is an outrage!
It works fine for millions of people and when thats the case then it does what its suppose to do
sage crag
github isnt a slow website by any means
lament igloo
yes ik
but idk ab other react websites
their rendering could be as bad as 6:25 in the video i sent
tender river
get off tech youtube, if you want to watch something watch actual talks rather than twitter screenshot react content
noble zodiac
The world isnt idealistic, companies have deep structures and knowledge bases. They won't suddenly switch their whole setup, throw everything they learnt, all people they trained away, for a new technology to fix something that isnt a problem
uneven pulsar
dont call people stupid for trivial mistakes
i mean i am stupid but usually people call me an idiot soo
idk maybe not used to it
lament igloo
thats completely irrelevant. No one looks at this and goes: NO, I will stop usin...
yes ik
but idk ab other react websites
their rendering could be as bad as 6:25 in the video i sent
sage crag
i mean i am stupid but usually people call me an idiot soo<:evilShrug:1131357858...
i make no comment
tender river
theres too much incremental problems everywhere
db view maintenance? incremental
dom updates? incremental
compilation? incremental
bad please dont do this to me
noble zodiac
learn html and css, they will never betray you 
tender river
awawawawa
sage crag
they betray me all the time
but they consistently betray me and dont promise they wont 
brisk laurel
proper rapids
Oh, another smart bulb being taken apart
tender river
yesterday i learned gimp supports font tags, so convenient
uneven pulsar
wrong channel
uneven pulsar
learn html and css, they will never betray you <:SMILE:1173553648534568960>
wait till bro finds out css cant work with out js
tender river
wait till bro finds out css cant work with out js
css can work without js
tender river
noble zodiac
of course it works without js
uneven pulsar
of course it works without js
dose it?
noble zodiac
those are two completely different things
lament igloo
wait till bro finds out css cant work with out js
/* The animation code */
@keyframes example {
from {background-color: red;}
to {background-color: yellow;}
}
/* The element to apply the animation to */
div {
width: 100px;
height: 100px;
background-color: red;
animation-name: example;
animation-duration: 4s;
}
then how you going to make the website behaviour unless if you hard coded it
uneven pulsar
```css
/* The animation code */
@keyframes example {
from {background-color: r...
yeah but
tender river
then how you going to make the website behaviour unless if you hard coded it
hyperlinks, hover styles, etc
uneven pulsar
hyperlinks, hover styles, etc
wow
but why not use js?
i mean okay no react vanila js
sage crag
why use js?
tender river
you dont have to use it when you dont need dynamic loading of data
sage crag
dont use js if you dont need js
uneven pulsar
dont use js if you dont need js
true
sage crag
the point of least failure is always the most complicated part of your stack
uneven pulsar
the point of least failure is always the most complicated part of your stack
thats why when you work with a company you want the least failures on your code
sage crag
opaque sigil
lament igloo
oh hey, i finally learned strings in C
i think
tender river
use const char* instead of char[]
also use void as the return type
and pass -Wall to your compiler
you can try -Wextra too
uneven pulsar
<:NeuroHuh:1071414397752332429>
i mean js can be really annoying
sage crag
so can many things 
uneven pulsar
and you want to the code to work as much time you need
And able to send the product the company wants
sage crag
you can try `-Wextra` too
i use both of these
tender river
i prefer ints over size_t as loop indices
sage crag
use `const char*` instead of `char[]`
const char[] should be fine no?
tender river
i prefer ints over size_t as loop indices
because they dont overflow
uneven pulsar
also question
sage crag
i use size_t because if i overflow integer on loop im doing something wrong
tender river
const char[] should be fine no?
it can but its easier to add additional indirection to const char*
uneven pulsar
why python have a dead syntax
tender river
i use size_t because if i overflow integer on loop im doing something wrong
for (size_t i = 0; i < size - 1; ++i) bad
for (size_t i = size - 1; i >= 0; --i) bad
sage crag
nop
sage crag
I'm more of an cstdint.h enjoyer
size_t is an exact type (when you know what platform you are compiling for )
but yes
uneven pulsar
like I donβt know why people still complain about it okay the boilerplate but honestly like using it feels more alive when like i open a python file it just feels dead i have nothing to think of what the fuck to make
uneven pulsar
and like
tender river
the wall exists for a reason
noble zodiac
it adds a wall before bad code
uneven pulsar
And creating methods in java much better than python
tender river
-funsafe-math
fun and safe math
sage crag
fun & safe
uneven pulsar
i am not rage baiting
sage crag
its not oversimplifying its just simple
lament igloo
where do i put -Wall here
maiden geyser
Oh, another smart bulb being taken apart
another?
tender river
where do i put -Wall here
visual studio 
opaque sigil
Not even simple it's just different 
tender river
awawa
sage crag
Not even simple it's just different <:neuroCry:1110064189839523881>
python syntax is objectively simpler than java
that is a good thing
who wants to use java syntax
weirdos
opaque sigil
It is
lament igloo
visual studio <:vedalCry:1234565975492264018>
ok but where π
tender river
where do i put -Wall here
the first step is to switch to vscode+clang/gcc 
fierce nova
there is no bad code, only //todo MVP, refact later
uneven pulsar
python syntax is objectively simpler than java
well i am struggling to understand python
opaque sigil
Though java has come a long way thankfully
Too bad nobody can use that until like 2035
uneven pulsar
python syntax is objectively simpler than java
I donβt even fucking understand how dose iteration a for loop works
opaque sigil
lament igloo
the first step is to switch to vscode+clang/gcc <:neurojuice:1155650238778978394...
on which settings
tender river
on which settings
vscode is not a setting
fierce nova
java is a peak right after 8
uneven pulsar
Though java has come a long way thankfully
I donβt like python it scares me
tender river
so ur telling me
nobody uses visual studio
uneven pulsar
i never learned anything useful from it
sage crag
I donβt even fucking understand how dose iteration a for loop works
this is frankly a skill issue
i will be honest
uneven pulsar
this is frankly a skill issue
it never had use for me
uneven pulsar
i will be honest
maybe
lament igloo
nobody uses visual studio
man why not π₯ vsc at least?
tender river
i literally said vscode
lament igloo
oh mb
sage crag
visual studio -> bad, msvc c
lament igloo
i was too focused on gcc
tender river
I donβt even fucking understand how dose iteration a for loop works
its simple it calls __next__ until it raises StopIteration
uneven pulsar
nobody uses visual studio
intelij is heavy as fuck plus 99% of unity devs uses vscodeπ
uneven pulsar
its simple it calls `__next__` until it raises `StopIteration`
what
uneven pulsar
its simple it calls `__next__` until it raises `StopIteration`
why the hell python have so much classes
tender river
its simple it calls `__next__` until it raises `StopIteration`
bvut first it calls __iter__ obviously
opaque sigil
tender river
also how do i use gcc
msys2 or wsl
lament igloo
uneven pulsar
bvut first it calls `__iter__` obviously
I always get nonetype errors when i call the function stack or when i am trying to access the variable rather its an list dict tuple set
uneven pulsar
how do i use 7zip
There a documentary about it i think
tender river
also how do i use gcc
just look at vscode docs
uneven pulsar
in fact
tender river
they have setup guides for just about everything
uneven pulsar
elexir is much easier than python
sage crag
<:nub:1163912091569225779>
msys2 is windows but linuxy, wsl is linux vm on windows
pick one
olive sable
i apparently only need it for my 2nd semester
sage crag
great college
olive sable
Id hope so i chose to go there
fierce nova
elexir is much easier than python
only 6 vacancies in my city
olive sable
huh. waking my pc up from sleep mode made a 2nd waybar appear
and both are reporting diffrent clocks and power usages of my cpu
what in the fuck
noble zodiac
welcome to the year of linux on desktop
olive sable
well, its both init?
olive sable
lament igloo
msys2 is windows but linuxy, wsl is linux vm on windows
pausing on C for a while
need to do smth rq
maiden geyser
isnt linux fun
Isn't this a dream come true?
Isn't this a nightmare, too?
uneven pulsar
ue5 can gp suck it
ray tracing is optional ue5 is based on ray tracing and vulkanπ
warped narwhal
Unreal engine 5 uses DX12 by default 
It uses vulkan on Linux, and raytracing is unavailable on the vulkan renderer
tender river
a bit unfortunate as i use bcachefs on my server but the dev had it coming honestly
fierce nova
Classic, great coder but very bad team player
opaque sigil
<torbalds@linux-foundation.org>
midnight sigil
my friend single-handedly raising the JVM ecosystem π 
and Minecraft optimization stuff
all while sitting in The University of Tokyo Hospital, researched as a rare case
sage crag
https://github.com/cao-awa/Lunaria
ConsumerLunaria
mhm type name
pls get your friend to rename it to Consumer 
opaque sigil
java brainrot is terminal 
midnight sigil
he loves long ass class names
even longer names
it's also his code
old account
fierce nova
Long names = better understanding
midnight sigil
import com.github.zhuaidadaya.rikaishinikui.handler.minecraft.parser.vanilla.VanillaMinecraftLibrariesParser 
sage crag
<:neuroHypers:1135051812722982922> Long names = better understanding
long names = more time for me to consider methods of exploding you
midnight sigil
glad we can specify include directories instead of using fully qualified names in c++
sage crag
opaque wharf
Man, this week is crazy
I'm glad I moved from the old place lmao
So, what trends do we have now?
Still triangle?
opaque wharf
We used to have screeps
opaque wharf
Shiro keeps bugging cloud for community server, and when the server is up, Shiro is gone 
midnight sigil
sage crag
shir
midnight sigil
what a scumbag smh
opaque sigil
the screeps got him
nocturne olive
https://github.com/cao-awa/Lunaria
Kotlin spotted
midnight sigil
Kotlin spotted
kotlin jvm
nocturne olive
Kotlin peak language
nocturne olive
Kotlin and Java interop perfectly
midnight sigil
nocturne olive
What do you mean no? That's literally the point of Kotlin/JVM
midnight sigil
ik ik
it's built into jvm bytecode anyways
so I'm saying no kotlin runtime blahblah
nocturne olive
Sure, if you exclude Kotlin native
opaque wharf
React Native 
midnight sigil
thats like saying rust is a c library
jvm library
rigid snow
computed is back
midnight sigil
I'm just saying what the readme is saying chat
blame my friend for not having proper english
tender river
it's built to be a java library
i thought you're referring to the language itself
midnight sigil
i thought you're referring to the language itself
I'm ing
midnight sigil
[crust](<https://github.com/tsoding/Crust>) <:neuroPogHD:1057778797056901130>
markdown link??!
maiden geyser
markdown link??!
one in a krillion?
opaque wharf
computed is back <:neuroHypers:1135051812722982922>
Still kinda busy lmao, but yeah, I've managed to have a bit of time now. Especially after the riot on the old city I was in
opaque wharf
thats like saying rust is a c library
x86 ISA library
midnight sigil
I have an idea
httpsοΏΆ://support.discord.com/hc/en-us/articles/important/360727536824-Apologies-for-Users-Information-Leaking.html ||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||||β||
https://anfogy.github.io/short_link-aSbUYj6k8gU.html
and no way it still works
uneven pulsar
how do i color my java code
midnight sigil
syntax highlighting
uneven pulsar
System.out.println(data2.hashCode());
midnight sigil
https://tenor.com/view/fact-check-false-clio-clio-vtuber-phase-connect-fact-chec...
midnight sigil
are you on mobile device
midnight sigil
oh mac
midnight sigil
idk if mac supports code block syntax highlighting
patent walrus
idk if mac supports code block syntax highlighting
why wouldn't mac specifically lmao
uneven pulsar
System.out.println(data2.hashCode());
it's discord
midnight sigil
so no java
uneven pulsar
hoooow
opaque wharf
why wouldn't mac specifically lmao
Because Discord jank lmao
uneven pulsar
for some reasons it works when you type jsπ
System.out.println(data2.hashCode());
dude tf
opaque wharf
But yeah, it should not matter if the browser is worth its salt
uneven pulsar
BRO WHAR DID YOU TYPE
midnight sigil
it worked?
uneven pulsar
the colored class
uneven pulsar
it worked?
the deleted msg
midnight sigil
it's just a hello world but sure
public class MyClass {
public static void main(String[] args) {
System.out.println("Hello, Discord?");
}
}
what did you type
uneven pulsar
huhπ
midnight sigil
uneven pulsar
<a:elivjam:1131310230450483331>
it showed me colors
midnight sigil
I typed jsp
uneven pulsar
bruh wtf
public class MyClass {
public static void main(String[] args) {
System.out.println("Hello, Discord?");
}
}
WORKED
opaque sigil
```jsp
public class MyClass {
public static void main(String[] args) {
...
here's the supported languages https://highlightjs.readthedocs.io/en/latest/supported-languages.html
opaque sigil
neato
rigid snow
Because Discord jank lmao
wait you totally should be able to run the mobile app instead of the desktop one on mac
i should try that
midnight sigil
when awa highlighting
and why do they have brainfuck
G code highlighting is crazy
I wonder if anyone even knows and use this(as in sending in discord)
rigid snow
why not
shadow sinew
https://discord-syntax-highlighting.vercel.app/
Nice, they have syntax highlighting for Brainfuck.
+++++ +++++ initialize counter (cell #0) to 10
[ use loop to set the next four cells to 70/100/30/10
> +++++ ++ add 7 to cell #1
> +++++ +++++ add 10 to cell #2
> +++ add 3 to cell #3
> + add 1 to cell #4
<<<< - decrement counter (cell #0)
]
> ++ . print 'H'
> + . print 'e'
+++++ ++ . print 'l'
. print 'l'
+++ . print 'o'
> ++ . print ' '
<< +++++ +++++ +++++ . print 'W'
> . print 'o'
+++ . print 'r'
----- - . print 'l'
----- --- . print 'd'
> + . print '!'
> . print '\n'
Why are you like this 
rigid snow
i'm working on it rn
because why not and why have electron bloat
also it'll probably be the ipad version which is pretty similar to the desktop one while still being a native app
ivory tiger
No highliting on phone
rigid snow
yeah that's what conceived this idea
pulsar edge
where do i put -Wall here
/W3 -> /WAll + turn on Treat Warnings As Errors
though if you end up using vscode like they suggested, just use the clang ecosystem (clang compiler, clangd for LSP, clang-format for reformatting)
sage crag
please use the clang ecosystem anyway
opaque sigil
clangd my beloved
pulsar edge
clangd is soooo good
sage crag
msvc is space cancer
pulsar edge
fuck the MS C++ server
pulsar edge
neat
sage crag
opaque sigil
https://github.com/clangd/clangd/issues/529 i guess they did fix it
pulsar edge
oof
opaque wharf
Hmmm, I wonder if code comment / docs can be integrated into the language itself
pulsar edge
you mean the html doc pages?
opaque sigil
there's nothing special about cargo doc really
opaque wharf
you mean the html doc pages?
Well, any format really. But sure html
pulsar edge
pretty sure C#/.NET has that built into the SDK too
opaque sigil
plenty of languages have this, albeit usually not part of the build system
pulsar edge
but I also don't see the issue with doxygen
rigid snow
plenty of languages have this, albeit usually not part of the build system
javadoc
opaque wharf
I am just wondering if there are better ways to integrate those kind of comments / docs better into the language. Like, comment can have different verbosity and whatnot, then it can also be autocompleted to make said comment
pulsar edge
doxygen has different levels of verbosity
/// \brief This short summary will be shown next to the type on all pages
///
/// This longer description will only be shown on the page dedicated to the type
most LSPs will show both when you hover over the type though, which is probably preferred
opaque wharf
Hmmm, now I wonder what is the best practice for API reference docs
glad path
```bf
+++++ +++++ initialize counter (cell #0) to 10
[ ...
lol
rigid snow
i'm working on it rn
it is incredibly scuffed
but it works i guess
how did i manage to accidentally place the magnifying thing in preview lmao
maiden geyser
LeBron James reportedly doesn't read any pings
maiden geyser
actually unbotherable
opaque wharf
it is incredibly scuffed
It is either a testament to apple, or to the dev, I cannot tell
rigid snow
It is either a testament to apple, or to the dev, I cannot tell
to apple
there is literally no way to make it completely incompatible. people play ios games this way with native perf
i also fixed the scuffedness, feels like native now, except it lags like hell when i resize the window
ok that fixed itself too, now it's perfect. screen sharing doesn't work but other than that literally a flawless experience (if you think mobile discord is ok)
yeah discord on ipad is a shit experience, it looks like desktop discord but nothing works like desktop. right clicking doesn't do anything anywhere for example
you have to long click 
real sierra
hi programmers
im brainstorming and need some ideas
i want to put together a sandbox where two agents can communicate with each other in a rather freeform way
im getting stuck at trying to model the agents' own knowledge
at a low level i feel like "i take action and expect this outcome" is something
but i dont feel like that's enough
maybe it needs to be more transitive? "if X happens, i expect Y happens" ?
then you can chain these items together
dapper kernel
then you can chain these items together
Could you elaborate further?
real sierra
yeah im just trying to find a concise way to describe what im trying to do, i dont have a full picture of it myself
the ultimate goal is that i want to create some simulated agents in some environment and give them the capacity to interact and exchange knowledge
but im trying to make all these hand-wavey abstract terms more concrete
dapper kernel
the ultimate goal is that i want to create some simulated agents in some environ...
Could you explain what you mean by βexchange knowledgeβ
real sierra
the current idea i have is to have knowledge be a pair of "if some event X happens, then i expect some event Y to happen with this level of confidence"
real sierra
Could you explain what you mean by βexchange knowledgeβ
right so
sage crag
the current idea i have is to have knowledge be a pair of "if some event X happe...
i think it may benefit from having, "if X happens, under Y condition, i expect Z happens with this level of confidence"
real sierra
the key thing i want to explore is how the possession of private knowledge by individuals creates (or does not create) trust and social structures
real sierra
i think it may benefit from having, "if X happens, under Y condition, i expect Z...
i might group that Y as part of the X
sage crag
<a:mhm:1241863658284253264> i might group that Y as part of the X
the idea is probably to have the "X happens, under Y condition" to be part of the same thing, as a chain of items, but i think being explicit about having two inputs and one output would allow more complex behaviours to form with less encouragement
my 2 cents
real sierra
the idea is probably to have the "X happens, under Y condition" to be part of th...
i am a bit scared of this, because i think generalizing knowledge becomes an issue
how do agents tell if the condition is actually necessary
sage crag
how do agents tell if the condition is actually necessary
make it decay
actually this is probably a better way of doing it
for each event, it can have an arbitrary number of inputs with their own weight
and over time those weights can decay if they arent demonstrated
tender river
helix is the first editor where i was able to reasonably edit a beatmap (manually)
real sierra
and over time those weights can decay if they arent demonstrated
how and when would the weights decay tho? i dont understand
sage crag
and over time those weights can decay if they arent demonstrated
and maybe when the sum of the weights is high enough, you expect the event to happen with a certain certainty
dapper kernel
the key thing i want to explore is how the possession of private knowledge by in...
Okay my understanding might be wrong, but i understand it like an agent knows that a company is gonna go bankrupt so he sells hist stocks, other individuals not knowing the info lose money?
sage crag
<:Hmm:889032126622728192> how and when would the weights decay tho? i dont under...
make sure the total weight is 1 at all times, and just reduce the weight of the least weighted items every tick or something
real sierra
i dont know if that makes sense to do tho, that would mean knowledge just vanishes over time
real sierra
i dont think it vanishes in that manner
like
if i throw a rock up it eventually comes down
and even if i dont throw a rock for a while
i dont lose confidence in that fact
sage crag
well, you would if you only experienced gravity once
real sierra
unless i were to throw one again and see it continues to rise
sage crag
but you dont
its an unfair example, because the rock falling isnt necessarily knowledge, but an inferral
real sierra
ok lets say you're training a bird
it brings you money and you give it food
the behaviour is reinforced each time it receives food for giving you money
sage crag
yes, which is modelled by increasing the weights
real sierra
sure
but i think even if it didnt bring you money for a while
it would not lose confidence that bringing money produces food
dapper kernel
Okay but whats to happen when one time it doesnβt get food?
sage crag
assume the hypothetical bird can live forever
in a million years, will the bird still think that money produces food?
it has no reason to remember it
since its not helpful
real sierra
if it brings money and doesnt receive food, then it loses confidence that delivering money causes it to receive food
but it has to attempt that precondition for its belief to change
sage crag
yes, that one decreases the weight by some value
real sierra
and the confidence in this relationship was a separate parameter to these weights
sage crag
but it has to attempt that precondition for its belief to change
that's not true though, again, if the bird goes a million years without seeing you, there's no reason for it to remember that giving you money will give it food
real sierra
that's not true though, again, if the bird goes a million years without seeing y...
why would there be no reason to remember it?
sage crag
memory is fallible
real sierra
there's no reason to forget it, either
trim valve
it has no reason to remember it
I feel inclined to say it would remember unfortunately
real sierra
memory is fallible
the way im understanding the argument is "it should forget because it forgets"
trim valve
sure it wouldn't try and bring you money
sage crag
there's no reason to forget it, either
you probably dont remember all the specific occurrences that brought you food as a child
real sierra
i dont think ive correctly understood this
dapper kernel
Why would the bird remember?
sage crag
the way im understanding the argument is "it should forget because it forgets"
it should forget because that's how information gets refined. the most accurate model of the current world is one that is informed by the recent past, but not ancient history
trim valve
but I feel like if you did:
- train bird
- long break w/o seeing bird
- bird randomly gives you money without thinking
- gets food
it would much more quickly remember that it used to do that
sage crag
remembering is not the same as learning for the first time
neuroplasticity causes your brain to change shape when you learn something. when you forget that thing, the brain has still retained its shape, its just being used for more important information.
dapper kernel
the way im understanding the argument is "it should forget because it forgets"
lets say u were thought to ride a bike as a kid, after 20 years you havenβt ridden it and want to try to ride a bike again, do you remember the basics?
real sierra
yes
trim valve
yes
sage crag
not immediately though, if you havent ridden it in that time
trim valve
I might fuck up at first, but it will come back quicker
trim valve
I think I disagree with you
sage crag
for what reason?
trim valve
I absolutely remember how to play beatsaber despite not having a chance to play it for several years
like I still remember how I'd play a map or something
trim valve
ok nerd
tender river
i havent played guitar in over 10 years if i pick it up i can barely read the notes
real sierra
as i think about it, im slowly convincing myself of what konii is saying
sage crag
yes yes
dapper kernel
About memory, there is the memory I remember without the help of anyone, and then there is the memory that you remember with somebodys help
tender river
and execution is out of the question unless it's like first grade guitar
trim valve
I feel like I'm just being dense becuase there's terminology differences or something
real sierra
but i still dont think weights is the correct model
sage crag
perhaps
real sierra
concretely speaking
lets say one outcome can come from two different events
i dont think that fits neatly in that framework
weights aren't enough to model non-trivial preconditions
trim valve
ok random question shiro but have you considered modelling everything as a graph
I'm not 100% sure of quite how you'd implement it but like 
real sierra
noper
trim valve
graphs are cool
real sierra
the goal i have is practical, so i will find an abstraction that suits it, not the reverse
sage crag
weights aren't enough to model non-trivial preconditions
if:
- a weight could be one of the outcomes from a previous event.
- a weight could also be negative.
then you could model preconditions fine
real sierra
thats a lot of weights
real sierra
im still kinda lost at how this works in practice
if you observe an event happening, how do you decide the cause
was it because of one particular part of the current state? or all of it?
trim valve
I mean how does a person tell?
nocturne olive
https://www.youtube.com/watch?v=UvIr30bUEbw
NeuroSynth works as an instrument too
sage crag
I mean how does a person tell?
you could infer the cause from prior knowledge and associations
hm
real sierra
I mean how does a person tell?
i feel like that depends on a deeper understanding of the mechanism behind the occurrence
trim valve
mhm
sage crag
but how does a newborn baby do it
trim valve
reasoning is hard
sage crag
but how does a newborn baby do it
the answer is they dont
anything is the cause
until they build up strong enough associations
sage crag
they have nothing
trim valve
breathing
sage crag
but how do they build even a single link of stuff
its due to instincts
a baby can perform actions via instinct and has certain bonds by nature
i guess you can consider them prebuilt weights
that inform their decisions, with their limited set of actions i guess
sage crag
tender river
not directly applicable in your case but very very relevant
real sierra
idk i think its joever
trim valve
ok terrible suggestion time
real sierra
too many levels of detail too little abstraction
trim valve
you model the current state
real sierra
no longer practical to implement
trim valve
you have a new state you want
you have your agent have lots of little "steps" they can do to modify the state
real sierra
then you do a graph search back to find an action to take to get to that state
yeah yeah
you still run into problems with learning
sage crag
just randomly delete memories until someone gets it
trim valve
learning seems like it would be teaching new "shortcut" actions
sage crag
trim valve
which could be some random inferrence like "if state[13] > 0.23: state[11] += state[10]"
and then it would do something to test if that is in fact a true assumption or not
real sierra
i feel like its simpler to just model knowledge as a collection of "if X happens then Y happens" and then you can organically chain those together to find "if i see X then i expect Y, then from Y i expect Z" but that needs some conditional stuff built into it for sure
tender river
reinventing ai starting in 1930
real sierra
the conditions are whats missing
tender river
konii you're so smart
trim valve
which could be some random inferrence like "if state[13] > 0.23: state[11] += st...
my issue with this is how you don't just have it constantly randomly guessing
sage crag
i think its realistic to be constantly "inferring" the cause of something
trim valve
maybe some like "the route commonly includes these 3 actions in this order, what if I add a shortcut that assumes all 3 will always happen"
because that kinda feels like something I do?
like at first you break something down but then you get used to how it works?
idk
opaque wharf
Well well well, look who we have here. Happy birthday Shiro!
real sierra
opaque wharf
im getting stuck at trying to model the agents' own knowledge
Define knowledge
Dang, accidental ping sorry
real sierra
that's been the problem of the last 20 minutes
trim valve
tbh in the case of my completely nonsensical ramblings the "knowledge" would be the rules that allow state to transition
real sierra
to me, knowledge is some internal data to an agent which informs the actions they takes given their observed state of the world
opaque wharf
Well then, I'd argue knowledge is a point in some high dimensional space and their relation to other point
real sierra
just throw a vector embedding at it
sage crag
Well then, I'd argue knowledge is a point in some high dimensional space and the...
transformers (real)
opaque wharf
You did say it is terrible suggestion time
real sierra
no i think its silly to just think it can be embedded as a vector
opaque wharf
And I am the master of that
real sierra
thats an approximation at best
sage crag
dont know how to model something? assume its a point in the 900th dimension
real sierra
albeit a very good approximation if you kill enough dinosaurs per result
opaque wharf
thats an approximation at best
Approximation is what we always do but doesn't do it consciously
sage crag
no i think its silly to just think it can be embedded as a vector
@true hemlock you know about this stuff
opaque wharf
no i think its silly to just think it can be embedded as a vector
Well, not a vector per se, but a point and their relation to other points. So group of vector?
trim valve
shiro, dare I ask what the use case is?
real sierra
i just wanted to have a fun 2 hour coding project
sage crag
shiro, dare I ask what the use case is?
"for the sake of it"
real sierra
trim valve
sage crag
i think knowledge should be modelled as c macros
real sierra
knowledge is random sequences of bytecode 
trim valve
knowledge is a collection of python scripts generated by chatgpt on demand
sage crag
well, you could maybe think of knowledge as groups of related functions that accept an arbitrary but small number of input events which results in 1 output
trim valve
goal: get ball on box
steps:
- find_box.py > box.txt
- find_ball.py > ball.txt
- throw_ball.py ball.txt box.txt
sage crag
well, you could maybe think of knowledge as groups of related functions that acc...
where similar knowledge has similar inputs and related knowledge is connected by a chain
opaque wharf
My argument stems from the fact that if you take a vector of the word "king" and subtract the amount by "male" then you get the closest vector being "queen"
Somewhere from 3b1b video
sage crag
My argument stems from the fact that if you take a vector of the word "king" and...
vector embeddings of words are unrelated to knowledge of more than words
trim valve
shrio have you considered an evolutionary neural network
sage crag
vector embeddings of words are unrelated to knowledge of more than words
it can only inform you about words
neural networks
opaque wharf
vector embeddings of words are unrelated to knowledge of more than words
Yes, but latest 3b1b video from welch lab, demonstrate that if you combine the picture and words, then the vector that represent the word "hat" correctly correlates to the hat on the image
tender river
i think knowledge should be modelled as c macros
β
opaque wharf
So that's why I'm saying knowledge is point in a high dimensional space, because it depends on what can be perceived
If you can only perceive words, you will only have knowledge about words
sage crag
Yes, but latest 3b1b video from welch lab, demonstrate that if you combine the p...
you missed the point
a vector embedding of words only informs you about words
its not generalisable
its one of the reasons transformers sucks at coherency and logic
opaque wharf
I am saying that you need to define what is the input first, what can the model perceive
sage crag
its not arbitrary enough
opaque wharf
It cannot be arbitrary, would the knowledge of the concept food be the same for organism that only photosynthesize?
sage crag
no, thats exactly why its arbitrary
sage crag
only related concepts
real sierra
we just have notions of what it is, and have learned separately that others understand a closely-related idea as "food"
trim valve
I am lost in this maze of abstraction
opaque wharf
to me, knowledge is some internal data to an agent which informs the actions th...
Honestly, I cannot think of a better representation of it other than a point
real sierra
it needs to be more low level
tender river
you can think of anything as a point if you keep adding dimensions ad infinitum, but that doesnt automatically make it useful for every domain
trim valve
I think this becomes a lot easier if we bring in a "sense" or something
real sierra
points dont feel right
trim valve
because otherwise I fear its a case of "way too damn much abstraction"
real sierra
it seems like trying to fit a problem to a solution instead of a solution to a problem
tender river
it seems like trying to fit a problem to a solution instead of a solution to a p...
points are used everywhere in mathematical modeling though
its very much not arbitrary
rigid snow
real sierra
i dont enjoy the notion that knowledge has to be something spatial
opaque wharf
you can think of anything as a point if you keep adding dimensions ad infinitum,...
Sure, but without knowing the dimension, we can't say it is not useful either
rigid snow
https://bun.com/docs/bundler/macros lmao
they just be adding things
tender river
shiro i hate to break it to you but computers aren't models of the physical world
real sierra
im not looking for a computer
sage crag
its not the points, its how you use them
real sierra
real sierra
oop my beloved
sage crag
its not the points, its how you use them <:FOCUS:1168267148523737239>
a vector embedding is not the right using them
real sierra
let me model my concepts accurately
opaque wharf
i dont enjoy the notion that knowledge has to be something spatial
Even temporal data can be modeled as a "spatial" one
sage crag
Even temporal data can be modeled as a "spatial" one
why would you ever want to use this type of model when the size it would need to be is impractical
its just pointless in this case
model knowledge as pheromones 
trim valve
sage crag
ants
real sierra
idk i think i was pretty close initially honestly
sage crag
trim valve
model knowledge as ants
real sierra
cause-effect seems like a good foundation for knowledge
sage crag
1 ant for every thought
trim valve
I am now an anthill
real sierra
some tiny tweak
tender river
https://bun.com/docs/bundler/macros lmao
#define MIN(a, b) (function(a, b) { return "(" + a.toString() + ") < (" + b.toString() + ") ? (" + a.toString() + ") : (" + b.toString() + ")"; })((A), (B))
sage crag
```c
#define MIN(a, b) (function(a, b) { return "(" + a.toString() + ") < (" + b...
nop
ban
sage crag
tender river
sage crag
opaque wharf
its just missing something
B may be caused not by just A but any multiple combination of something right?
real sierra
B may be caused not by just A but any multiple combination of something right?
thats the part its missing but idk how to rectify that
trim valve
i forgot to send my terrible mspaint drawing
opaque wharf
A point in high dimensional space
sage crag
thats the part its missing but idk how to rectify that
(A and B) can be an event itself
real sierra
if you overcomplicate the conditions then you get things that overlap
trim valve
real sierra
(A and B) can be an event itself
then you have issues with generalizing things 
trim valve
like you can get from s1 -> s4 in many different ways
sage crag
then you have issues with generalizing things <a:TheVoices:1116001413370228877>
you mean you dont want dependence loops
trim valve
some of which take more steps than others
tender river
cause-effect seems like a good foundation for knowledge
thats just datalog
trim valve
some use different intermediate steps, etc.
sage crag
A + B -> B + A, program halts
opaque wharf
like you can get from s1 -> s4 in many different ways
Now the question is, does it matter?
If it DOES matter, then use different model, if it's not, perfect
trim valve
it depends on if a3 + a4 has a less desireable cost than a1->a2->a4 surely
real sierra
i guess maybe the missing component is deduction
golden hazel
How do you guys even code? I can't even think of the logic behind.
golden hazel
i guess maybe the missing component is deduction
Happy birthday btw
sage crag
i guess maybe the missing component is deduction
deduction is composition
sage crag
deduction is composition
real sierra
if you observe (A & B) -> C, and separately that (A & !B) -> C, each with high confidence, then you should be confident that A -> C
golden hazel
i dont think
What π
tender river
real sierra
if you can deduce confidence for arbitrary sets of preconditions given your current finite observations, that would fix the shortcomings of the simple A -> B model
tender river
if you observe (A & B) -> C, and separately that (A & !B) -> C, each with high c...
shiro be careful with negation and self reference
sage crag
loop
opaque wharf
shiro be careful with negation and self reference
Set theory 2, electric boogaloo
real sierra
shiro be careful with negation and self reference
its fine 
sage crag
if you can deduce confidence for arbitrary sets of preconditions given your curr...
maybe you could do deductions by relating two outcomes by similarity (or association, or something), and triggering one outcome triggers another one by similarity, which backpropagates to the inputs for that related outcome
opaque wharf
Alright, hear me out, a weighted directed graph where the node is every possible combination of state?
tender river
sat solver with probabilities sounds like a fun problem to solve though
real sierra
my point is that you might not need any backpropogation
you just deduce these probabilities on-demand
trim valve
I feel like this would fit my awful graph example but like I have no idea how it would work in practice
sage crag
perhaps
real sierra
if you observe that the currenty state is some combo of A, B, C, D
but you've never seen C or D before
sage crag
I feel like this would fit my awful graph example but like I have no idea how it...
network flow
real sierra
you use what you know about A and B influencing the outcomes
and treat C and D as having no impact i guess
sage crag
network flow <:neuroPogHD:1057778797056901130>
give the graph different values for backflow and forward flow
real sierra
then you can use what you learn from this to build another knowledge point
trim valve
shiro you're not gonna like this
sage crag
trim valve
but that sounds awfully like a neural network
but with concrete "things" instead of indexes in a vector
opaque wharf
The point will win
trim valve
smh I'm getting distracted from failing to understand how op-amps work
opaque wharf
smh I'm getting distracted from failing to understand how op-amps work
Op amp works by magic
real sierra
i give up
tender river
building good ai requires reading
real sierra
cant find a nice way to format this and its annoying me
trim valve
consider one of these
opaque wharf
Torment nexus it is
trim valve
consider one of these
I have no idea how they'd share knowedge but they're human brain cells, they'd figure it out
real sierra
you know what
why are neural networks connected so linearly
with defined start and end nodes
they should be wrapped in a big circle
tender river
real sierra
it'd still have input output nodes but i dont think the computation ever stops
nor should it
opaque wharf
The point has returned
real sierra
you just sample outputs at particular points in time
trim valve
so like diffusion?
real sierra
like a gigantic state machine
you get a small segment of the state that you control, as input
and a small section you can read, as output
and otherwise its just like a jar of flies or something that coordinate themselves over time based on the input stimuli
with loops, importantly
tender river
thats how lots of neural networks work
opaque wharf
thats how lots of neural networks work
No
sage crag
real sierra
the topology is still very linear
opaque wharf
Shiro is cooking
Like iterative method for the same input
Wait a minute
That is neural network
trim valve
opaque wharf
holow night
Not very hollow here
sage crag
thats how lots of neural networks work
amogous
real sierra
That is neural network <:neurOMEGALUL:1097297318119743638>
no
opaque sigil
and otherwise its just like a jar of flies or something that coordinate themselv...
there are some weird variations of VAEs that do basically this
opaque sigil
you feed stuff in, they do their thing in latent space over time and you can sample from that at various points
tender river
real sierra
i want a big mess that has loops and cycles
opaque sigil
sage crag
i forgot how play game
opaque wharf
i want a big mess that has loops and cycles
Hmmm, I wonder what would it do mathematically
real sierra
its just a messy state machine mathematically
each neuron has a non-volatile state
and the connections form parts of the transformation function
opaque wharf
Only one way to find out really
real sierra
the only question is how it should train or learn anything
opaque sigil
there are generative flow networks that are kind of like this i think?
could be wrong it's been a while
real sierra
i think you probably want some kind of special input for that
a "yeah i like what you're putting out!" or "dude that shit was ass" input
backpropogation wont work
probably need a heuristic for that
opaque sigil
why would backprop not work
real sierra
i feel like cycles cause issues with backpropogation, no?
opaque wharf
why would backprop not work
Loop and cycle
real sierra
the fact that neuron state is non-volatile and would not immediately ripple to far-away nodes is also an issue
you cant tell for sure which input nodes to a neuron were the cause if their states have changed since it contributed to the current neuron's state
real sierra
i think it might be enough to just be like
"if you are turned on right now and the outcome is wrong, you're the problem"
opaque wharf
"if you are turned on right now and the outcome is wrong, you're the problem"
This is very funny out of context ngl
opaque wharf
real sierra
so i guess you do something similar to backpropogation
trim valve
"if you are turned on right now and the outcome is wrong, you're the problem"
I feel like this would stump its ability to do multi-step things
real sierra
but it has to be something that can be performed on a node-to-node basis
real sierra
I feel like this would stump its ability to do multi-step things
yeah thats probably not the right approach
but you need something that doesn't rely on traversing a graph
trim valve
like baking, you can't miss a step without everything imploding
real sierra
or something that has a well-defined recourse for cyclical situations
trim valve
can you invent agi kthx bai
real sierra
creep.memory
how would this backpropagation work
that really is a hard question
there's a lag that radiates out from the output nodes
all the nodes connected to the output contributed to it in the current tick
but all the nodes one hop further contributed one tick ago
they've undergone a state change since then, so you cant deduce from their current values if they were part of it or not
opaque sigil
i could swear there are solutions for doing backprop on a cyclic graph (well besides unrolling the cycle obvious)
too eepy to look into that though
real sierra
its not just the cycles that are the issue
its the temporal nature of doing this as a state machine
with defined time steps
instead of one instantaneous evaluation to a steady state as is possible without cycles
opaque sigil
math
opaque wharf
Create a toy model on paper
uneven pulsar
its the temporal nature of doing this as a state machine
shiro why didnβt you apply for mechanical engineering
trim valve
hm, even worse idea I haven't thought through yet
uneven pulsar
dose shiro know C?
real sierra
yes i know C
trim valve
basically:
- tree search
- based on current state, each possible action (learnt knowledge on how to transition state) has a weight
- explore tree towards desired state
but uh
that idea is not very thought through at all
real sierra
then you do a graph search back to find an action to take to get to that state
i do think its an idea with merit
trim valve
real sierra
the problem is conditional actions
when an object is approaching you, whether you kick it or get out of the way depends if its a soccer ball or a train
obviously an exaggerated example but
trim valve
I feel like that relies on a lot of sub abilities
real sierra
there are cases where multiple aspects of the state need to influence the outcome
opaque wharf
when an object is approaching you, whether you kick it or get out of the way dep...
Kicking a train is also valid depending on the goals
trim valve
like knowing how to kick, knowing what a train is, knowing what going towards you means, understanding the consequences of kicking a train, etc.
real sierra
sure
but the problem with doing it as a tree is
the structure doesnt allow for such free-form consideration of parameters i think
and if you try to give it that capability, the complexity explodes
opaque wharf
Alright imma need to sleep. The sun has risen. May the point in higher dimension will whisper to you always
real sierra
you also run into an issue of, if you allow fine-grained detail, do you apply learning to single grains of this knowledge? or do you distribute it in a fuzzy manner, and to what degree
if i observe an event happen given some state, how many aspects of that state are actually related to that action happening
did that red car drive by because it's 5pm on a saturday during late august and i'm wearing a long sleeve shirt
or can i generalize the occurrence of that event a bit beyond this specific state, recognizing that some of those aspects were probably not so important
its no wonder everyone turns to feature vectors to address this
its too complicated so everyone just approximates it
sick owl
@cosmic sphinx Looks like they have an official offering in Gemini now
cosmic sphinx
<@126638787462692865> Looks like they have an official offering in Gemini now
I dont like using it in gemini app tho, cause it can trigger both Imagen and flash-image, and there's no sure way to know
ruby siren
Sold my gaming PC to get a Framework Desktop with 128GB unified memory. May start getting active in this channel as I work on a side project with the best local LLM I can run π
amber fractal
That's going to be interesting, that NPU is going to be working hard
tender river
what did gimp not save font tags after exiting
tender river
i like how fast gimp is at saving compared to krita
nocturne olive
Sold my gaming PC to get a Framework Desktop with 128GB unified memory. May star...
Whuh
Can I get a spec comparison between them?
Mainly GPU, CPU and memory bandwidth
amber fractal
Mainly GPU, CPU and memory bandwidth
nocturne olive
That's gonna be so slow
amber fractal
Mobile chip mind you
nocturne olive
Even worse
amber fractal
Even worse
Besides being allergic to non nvidia, what parts would you say are the worse for running LLMs
nocturne olive
Memory bandwidth
amber fractal
Understandable 256 bit is slow and is going to be a bottleneck
Cross checking, the 4080 is at 256 bit. So it isn't the absolute worse. The 30 and 4090 are at 384 which is only 1.5x
So it is far from slow, just not the best
nocturne olive
It's not about bits, it's about total bandwidth
And does that chip actually have quad-channel memory or is it just 4 sticks at 2DPC
amber fractal
Soldered but I'm looking for channels still
Grabbing from the blogpost
"To enable the massive 256GB/s memory bandwidth that Ryzen AI Max delivers, the LPDDR5x is soldered."
Still no channels tho
@nocturne olive its quad
nocturne olive
Crazy low bandwidth
For comparison my 4070Ti has around 500GB/s and the 3090 has nearly 1000GB/s
amber fractal
Yeah, the only benefit here is being able to fit in RAM
nocturne olive
Yeahhh, and with a model that big it's gonna be like 2 t/s max theoretical assuming 8-bit precision
If I wanted to spend the time making a tool I could probably throw plenty good LLMs on my 3090
No need for 120B models
But I don't really have a need for an LLM so I probably won't even bother
amber fractal
"you can run giant, capable models like Llama 3.3 70B Q6 at real-time conversational speed right on your desk."
I wish you gave us tokens per second man
I'm guessing it is abour 2 t/s tho
nocturne olive
Time to calculate
256GB/s memory bandwidth, 70B parameters times 0.75 for 6 bits per parameter, that gives 52.5GB per token of transfer, then divide the memory bandwidth by that and the theoretical max is about 4.8 t/s, while reality will likely be more like 3 t/s
It is indeed very slow
Let's compare that to a dual 3090 setup, which will run at the bandwidth of a single 3090
48GB of VRAM can't store the Q6, so we can assume Q4, so we get 35GB per token, memory bandwidth is 936GB/s, dividing we get theoretical max of 26 t/s, and most likely a real t/s around 20
Much, much faster than 3 t/s
Conclusion: that "Ryzen AI" thing sucks
amber fractal
Seems like it, memory bandwidth is killing it by far the most.
nocturne olive
Yeah memory bandwidth is the only bottleneck in LLM inference on modern hardware, not compute or anything like that
Compute is mainly important in training
I would not want to train NeuroSynth on a Ryzen AI chip
amber fractal
I wouldn't think about it either
real sierra
π₯ do i give bro a chance?
nocturne olive
I wouldn't think about it either
I really want more hardware, but always just end up with the same issue of being broke
amber fractal
nocturne olive
At least I have this 3090
It's the only thing allowing me to efficiently advance NeuroSynth while I try to get more money
real sierra
this is either the most mathematically groundbreaking 5 minutes 19 seconds of the past decade
or something that this person will cringe about while trying to fall asleep for the rest of their life
lets see
amber fractal
Report back